reorg — local-model file reorganiser

SkillFiles & storage

Propose and, after approval, execute content-aware file reorganization or deduplication using a local Mac Mini model with a reversible undo record. Use when sorting a folder by file contents or finding exact and near duplicates. Always dry-runs first.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the reorg — local-model file reorganiser skill

What this skill tells your AI

The instructions your AI receives, as published by flonat/flonat-research in skills/reorg/SKILL.md and read by ahel’s review.

Conversational front-end for the local-reorg CLI (scripts/local-reorg/). The CLI runs a local Ollama model on the Mac Mini that reads each file's content and proposes where it should go, or finds duplicates. Nothing is ever moved without your approval, and every run is reversible.

When to use

  • "Reorganise / tidy / sort this folder" where you want the model to read files, not just pattern-match filenames.
  • "Find duplicates" / "dedup this folder" (exact + near-duplicate via embeddings).
  • General clutter (Downloads, Desktop, a messy project subfolder).

For research-project structural changes use an installed project-renaming or project-initialization workflow; if none is available, use a normal reviewed git mv plan with explicit backlink and configuration checks. For meeting to-sort/ inboxes, use the project's inbox-processing workflow. This skill is general-purpose file tidying.

How to invoke the CLI

First hostname:

  • On the Mac Mini ([server]): the reorg shim is on PATH — call reorg … directly.
  • Anywhere else (MacBook, etc.): run the installed Mini wrapper over SSH: ssh mini '~/.local/bin/reorg <args>'

The model + files both live on the Mini, so resolve the target using the Mini's Task Management path registries before invoking the wrapper. Do not embed either machine's physical Dropbox root. If you cannot resolve the path unambiguously, ask.

Workflow

  1. Confirm the target folder and pick a mode with the user if unclear:
    • reorg by scheme: auto (default) · by-type · by-topic · by-date
    • dedup: --dedup (report) → --dedup --quarantine (move redundant copies aside)
  2. Dry-run — never skip this:
    • reorg: reorg <folder> [--scheme X] [--recursive]
    • dedup: reorg <folder> --dedup The CLI writes <folder>/.reorg/plan.md (reorg) or .reorg/duplicates.md (dedup).
  3. Read the plan file and summarise it for the user — group counts, notable moves, and any ⚠low-conf items. Do not dump the whole file; give a scannable summary and the path.
  4. Get approval. The user may edit .reorg/plan.json first (trim/retarget moves).
  5. Execute only after a clear yes:
    • reorg: reorg <folder> --apply (add --yes if ≥20 moves — tell the user it's the break-the-glass threshold)
    • dedup: reorg <folder> --dedup --quarantine
  6. Offer undo: reorg <folder> --undo reverses the most recent applied/quarantined run (LIFO). Mention it after any execution.

Options worth surfacing

  • --model gemma4:e2b — ~1.5× faster than the default gemma4:e4b, slightly less accurate. Suggest it for large folders.
  • --max-files N — default 200; raise for big sweeps (warn about time: ~5–6 s/file on e4b).
  • --dedup-threshold 0.95 — stricter near-match (default 0.92; paraphrases ~0.96).

Safety (the CLI enforces; restate to the user)

  • Refuses Overleaf (Apps/Overleaf/), the vault (vault/), AI client homes, and anything under data/raw/ or .git/. Don't try to work around this.
  • Never deletes. Reorg moves into subfolders; dedup quarantines into .reorg/duplicates/.
  • git mv for tracked files; undo manifest on every applied run.
  • The dry-run → approve → apply loop is mandatory. Do not run --apply/--quarantine before showing the plan and getting a yes.

Requirements

Ollama must be running on the Mini with gemma4:e4b (or the chosen model) + nomic-embed-text for dedup, plus pdftotext. All present as of setup. If Ollama is unreachable the CLI exits with a clear message — relay it rather than guessing.

Signals

GitHub stars
133
Forks
24
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
reorg
Source
github.com/flonat/flonat-research